Re: fio replay

[Sitsofe Wheeler <sitsofe@yahoo.com>]

On Wed, Feb 26, 2014 at 03:48:54PM +0000, Lamkin, Eric wrote:
> 
> Were you replaying on the same host where the blktrace was recorded?

Yes I was.

> I am not.

So long as the target block device is big enough I would hope it would
not make a difference so long as you were using replay_redirect . It
definitely works for me using different devices.

> I'll try with no --iodepth command.  That is all I see different
> between your methodology and mine.

That should not make a difference other than making things slower. The
real command I'm using is
fio --read_iolog=sdd.dump --ioengine=libaio --iodepth=10 --direct=1 --thinktime=15000 --thinktime_blocks=16 --replay_no_stall=1 --replay_redirect=/dev/sdf --name=replay

> What is btreplay?  Does it allow device mapping if the traced devices
> have different names on the replay host?  Will btreplay run
> unconstrained by the original timings?

This is a bit off topic but I'll take stab at answering anyway.
btreplay "recreate[s] IO loads recorded by blktrace"
(http://linux.die.net/man/8/btreplay ). Usage is as follows:
blktrace -d /dev/sdd
btrecord sdd
btreplay sdd 

Device mappings are supported using the -M/--map-devs option. It can be
made to go as fast as possible by using the -N/--no-stalls option.

You may also want to look at blkreplay by Thomas Schoebel-Theuer of 1&1.

-- 
Sitsofe | http://sucs.org/~sits/